Environment International 73 (2014) 111–116
Contents lists available at ScienceDirect
Environment International journal homepage: www.elsevier.com/locate/envint
Carbon dioxide emissions and change in prevalence of obesity and diabetes in the United States: An ecological study Alexander R. Zheutlin a, Sara D. Adar b, Sung Kyun Park b,c,⁎ a b c
Program in the Environment, University of Michigan, Ann Arbor, MI, USA Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor, MI, USA Department of Environmental Health Sciences, University of Michigan School of Public Health, Ann Arbor, MI, USA
a r t i c l e
i n f o
Article history: Received 16 October 2013 Accepted 9 July 2014 Available online xxxx Keywords: Carbon dioxide Climate change Diabetes Greenhouse gas Obesity PM2.5
a b s t r a c t Recent studies suggest that increasing levels of the greenhouse gas, carbon dioxide (CO2), may influence weight gain and thus may play a role in rising trends in obesity and diabetes. We conducted an ecological study to examine the associations between CO2 emissions from fossil fuel combustion and changes in the prevalence of obesity and diabetes in the United States. County-level data on CO2 emissions, prevalence of obesity and diagnosed diabetes, other sociodemographic factors and neighborhood characteristics related to urbanicity, and fine particles (PM2.5) between 2004 and 2008 were obtained from the Vulcan Project, Centers for Disease Control and Prevention, and American Community Survey. Linear mixed effect modeling of 3019 counties for the associations between average CO2 emissions and changes in diabetes and obesity prevalence between 2004 and 2008 was performed. The average obesity and diabetes prevalence increased between 2004 and 2008 by 3.65% (SD: 1.88%) and 1.65% (SD: 1.70%), respectively. A marginally significant positive association between CO2 emission and changes in obesity prevalence was found with adjustment for sociodemographic factors, indicators of urbanicity and spatial autocorrelation (p-trend = 0.06). The association became weaker and nonsignificant with further adjustment for PM2.5 (p-trend = 0.17). There was a significant positive association between CO2 emission and changes in diabetes prevalence before controlling for PM2.5 (p-trend = 0.05) but the association became null after controlling for PM2.5 (p-trend = 0.49), suggesting that PM2.5 is a critical confounder in the association between CO2 emission and changes in diabetes prevalence. This study does not support the hypothesis that CO2 emissions, a leading driver of climate change, may be linked to increasing trends in obesity and diabetes, though there was an indication of possible link between CO2 and obesity. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Obesity and type-2 diabetes have been increasing dramatically over the past several decades in the world and can be considered global epidemics (Chen et al., 2012; Malik et al., 2013). According to the 2011 National Diabetes Fact Sheet, 36% and 11% of the United States (U.S.) population aged 20 years and older were afflicted with obesity and type-2 diabetes in 2010, respectively (CDC, 2011). Recently, the American Medical Association announced obesity as a disease (Hoven, 2013). It has been suggested that such epidemics are attributable to poor lifestyle behavior, such as higher caloric input and lower physical energy output. However, poor diet and sedentary lifestyle do not
⁎ Corresponding author at: SPH II-M5541, Departments of Epidemiology and Environmental Health Sciences, University of Michigan School of Public Health, 1415 Washington Heights, Ann Arbor, MI 48109, USA. Tel.: +1 734 936 1719; fax: +1 734 936 2084. E-mail address:
[email protected] (S.K. Park).
http://dx.doi.org/10.1016/j.envint.2014.07.012 0160-4120/© 2014 Elsevier Ltd. All rights reserved.
account for all of the variations in obesity and type-2 diabetes (Thayer et al., 2012), suggesting that other factors may also play an important role. A recent study by Hersoug et al. (2012) proposed that increases in obesity and type-2 diabetes may be due to increased atmospheric carbon dioxide (CO2). This hypothesis was motivated by a finding that animal species living under controlled environments (various laboratory rodents and primates) or near human establishments (cats, dogs, feral urban and rural rats) had significant weight gain over the last 50 years, although those animals have had the same diets for decades (Klimentidis et al., 2011). The documentation of greater average weight in numerous species indicates that there may be an environmental component applicable to all species impacting current trends in body composition. Plausible biological mechanisms by which higher CO2 exposure may lead to weight gain and other metabolic disorders are poorly understood, although a decrease in blood pH and subsequent hormonal reactions by elevated atmospheric CO2 were proposed (Hersoug et al., 2012; Sunanaga et al., 2009). To the best of our knowledge, there is no study examining the association between atmospheric
112
A.R. Zheutlin et al. / Environment International 73 (2014) 111–116
CO2 and obesity and type-2 diabetes. Any such association could have important implications since the subtle accumulative effects of climate change on human health are relatively unknown. Furthermore it may hint at a previously unconsidered cost of increasing CO2 in the atmosphere. Using data from the Vulcan Project, Centers for Disease Control and Prevention (CDC), and American Community Survey we examined whether there were relationships between average fossil fuel CO2 emissions and changes in diabetes and obesity prevalence at the county level in the U.S. between 2004 and 2008. We hypothesized that a positive change in the prevalence of obesity or diabetes would be associated with greater emissions of atmospheric CO2, independent of factors commonly associated with diabetes and obesity and neighborhood characteristics related to urbanicity, and fine particles (particulate matter less than 2.5 μm in diameter, PM2.5). We also examined a secular trend of blood CO2 as measured by serum bicarbonate levels in the U.S. using data from the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2012 to explore whether there is a temporal correlation between ambient CO2 concentration and blood CO2 in the general U.S. population. 2. Material & methods 2.1. CO2 emissions Data for annual CO2 emissions were collected from the Vulcan Project for years 2004 to 2008. The Vulcan Project is an initiative funded by the National Aeronautics and Space Administration/the U.S. Department of Energy that has quantified fossil fuel emissions with high resolution across North America (Gurney et al., 2009). Emissions were assessed based on million tons of carbon at the county level based on the U.S. Environmental Protection Agency's (EPA) National Emissions Inventory (NEI) data. The NEI provides carbon monoxide (CO) emissions by facility or unit, as well as nitrogen oxides (NOx) and emission factors (EF). CO, NOx, and EF are formulaically inserted to calculate a total fuel throughput, which is then used to find CO2 emissions by accounting for emissions from pollutant materials (Gurney et al., 2009). CO2 emission levels in the U.S. did not change much between 2004 and 2006, then increased in 2007 and decreased in 2008 (EPA, 2013a). Several factors influenced this year to year variability, such as energy (fuel) prices and weather conditions (EPA, 2013a). For the exposure metric of our study we used a five-year average of county level CO2 between 2004 and 2008. 2.2. Obesity and diabetes prevalence County-level obesity and diabetes prevalence between 2004 and 2008 was obtained from the National Diabetes Surveillance System at the CDC (CDC, 2012b). These data come from the Behavioral Risk Factor Surveillance System (BRFSS), which is an ongoing, monthly, state-based telephone survey of U.S. adults (CDC, 2012a). Interviewees were considered obese if their body mass index (BMI) ≥30 kg/m2, which was computed using self-report of height and weight. Interviewees who had been told by a physician they were diabetic were considered to have diagnosed diabetes. Women who indicated that they only had diabetes during pregnancy were not considered to have diabetes. The CDC used three-year averages of county-level estimates to improve the precision of data: for example, 2004 estimates were computed using data from 2003, 2004 and 2005 (CDC, 2013a). County level prevalence was reported as age-adjusted percent of the population aged 20 years or older. The primary outcomes were determined to be change in prevalence of obesity or diagnosed diabetes between 2004 and 2008, which was computed based on the difference in prevalence levels provided for 2004 and 2008. There are no temporal correlation issues in the outcome measures in the present study
because we examined the difference in prevalence as the outcome rather than absolute prevalence from multiple years. 2.3. County-level characteristics and risk factors Five-year estimates from the U.S. Census American Community Survey (ACS) covering 2005–2009 were used for county-wide covariate data (Department of Commerce, 2012). The ACS five-year estimate was selected due to greater precision and included more data points as opposed to one or three year estimates which included less counties. This included median age as well as percentages of males, persons aged 25 years or older with a high school degree or equivalent, and population with income below the poverty level. Poverty level is identical throughout the U.S. except Alaska and Hawaii, and is calculated based upon the household income, the household size, and the age of household members (ASPE, 2014). Median age and percent male are based upon all age groups within each county, as there is no available data for median age or percent male based solely on adults. The ACS offers multiple race/ethnicity identification methods, and for this study individual identification as black or African American, Asian (including Asian Indian, Chinese, Filipino, Japanese, Korean, Vietnamese, and other), white, or Hispanic or Latino was used. Other race/ethnic groups such as American Indian/Alaska Native and Native Hawaiian/Other Pacific Islander were not included because proportions are too low and some regression models including those variables did not converge. Population sizes per county were divided by county land area (given in square miles) in order to estimate population density. Density of fast food restaurants (limited-service eating-places per 100,000-population count) was obtained from the 2006 BRFSS Supplement Survey. Countylevel leisure time physical inactivity which was also based upon the BRFSS survey was obtained from the CDC's National Diabetes Surveillance System along with diagnosed diabetes and obesity (CDC, 2012b). Physical inactivity was defined as no participation in any physical activities or exercises other than regular job such as running, calisthenics, golf, gardening, or walking for exercise, during the past month (CDC, 2013a). County-level PM2.5, a combination of monitored and modeled data, was obtained from the CDC's National Environmental Public Health Tracking Network, a joint venture between the CDC and Environmental Protection Agency (CDC, 2013c). Additionally, variables were added to each dataset for the latitude and longitude of each county centroid to explore spatial trends. 2.4. Serum bicarbonate and ambient CO2 We obtained serum bicarbonate data from the NHANES cycles between 1999/2000 and 2011/2012 (CDC, 2013b). Bicarbonate concentrations were determined with Hitachi Multichannel analyzer (Model 917) in the NHANES 1999/2000 and 2001/2002 cycles; Beckman Synchron LX 20 in the 2003/2004 and 2005/2006 cycles; and Beckman UniCel DxC 800 Synchron in the NHANES cycles from 2007/2008 to 2011/ 2012. We also collected complex sampling design factors such as primary sampling unit, strata and sampling weight from each cycle. We restricted only adults aged 20 years or older. A total of 33,546 subjects were included. Annual averages of atmospheric CO2 concentrations at Mauna Loa, Hawaii were obtained from the Global Greenhouse Gas Reference Network available at the Earth System Research Laboratory, the U.S. National Oceanic & Atmospheric Administration (NOAA/ESRL, 2014). CO2 concentrations at the Mauna Loa Observatory represent the longest record of direct measurements of CO2 in the atmosphere since 1958, which covers CO2 data for the period of NHANES cycles between 1999 and 2012. 2.5. Data analysis Spatial maps were first created in order to present the distribution of CO2 emissions, diabetes, and obesity across the continental U.S. between
A.R. Zheutlin et al. / Environment International 73 (2014) 111–116
the years 2004 and 2008 using Arc GIS v9.3.1 (ESRI, Redlands, CA). We then examined the distributions (mean, standard deviation (SD), minimum and maximum) of all variables. We assessed the Spearman's rank correlations between covariates considered and the outcomes (changes in the prevalence of obesity and diabetes from 2004 to 2008) and the exposure (5-year average of CO2 emission). All data analyses were conducted using R version 2.15 (R Foundation for Statistical Computing, http://www.r-project.org). Of the 3140 U.S. counties and county-equivalents, 3019 (96%) were included for statistical analysis. The 121 counties excluded in the study were withheld due to missing data on fast food density, or percent of population above 25 years of age with a high school degree or equivalent. Additional 25 counties in Alaska and Hawaii where PM2.5 data are not available were excluded, leaving 2994 counties in the model with PM2.5. Mixed effects models were fitted to examine the relationship between changes in prevalence of obesity or diabetes and quintiles of CO2 emissions and to account for spatial autocorrelation. We used quintiles rather than continuous CO2 as our primary form of exposure because the relationships did not appear linear (supplemental Fig. 1). We considered three spatial correlation structures using information on the latitude and longitude of each county centroid; Gaussian, spherical, and exponential but selected exponential using the Akaike Information Criteria. Potential confounding factors considered include male prevalence, median age, race (percent black/African American, Asian, white, and Hispanic/Latino), poverty prevalence, high-school or equal educational attainment for those over 25 years of age, physical inactivity, fast food density, and population density. We additionally controlled for PM2.5 because CO2 emission was highly correlated with ambient PM2.5 concentrations and a recent study reported a significant association between county-level PM2.5 and diabetes prevalence in the U.S. (Pearson et al., 2010). For diabetes, we further adjusted for change in obesity. Parameter estimates and 95% confidence intervals (CIs) were computed by comparing each of the upper 4 quintiles of CO2 emission with the lowest quintile. Tests for linear trend were conducted using ordinal terms for the quintiles. We plotted biannual averages of ambient CO2 concentrations at the Mauna Loa Observatory (Hawaii) and serum bicarbonate concentrations against the continuous NHANES cycles from 1999/2000 to 2011/2012 to examine temporal trends and potential correlations. Survey linear regression was used to compute a biannual change in serum bicarbonate concentrations.
113
3. Results Obesity and diabetes prevalence have varied over the five-year span of 2004–2008, increasing by an average of 3.67% (SD: 1.89%, range: −3.4% to 12.6%) for obesity and 1.64% (SD: 1.67%, range: −6.2% to 9.5%) for diabetes (Table 1). The average CO2 emission level for counties across the U.S. was 0.55 million tons of carbon (SD = 1.29 million tons, range: 0.002 to 21.7 million tons). The county-level annual PM2.5 concentration averaged between 2004 and 2008 was 11.4 μg/m3 (SD = 2.47 μg/m3, range: 4.48 to 20.1 μg/m3). All sociodemographic variables, indicators of urbanicity, and PM2.5 were significantly associated with CO2 emission levels (p-values for trends across quintiles of CO2 emission b 0.0001). County-level estimates of the 5-year average CO2 emission levels and changes in the prevalence of obesity and diagnosed diabetes between 2004 and 2008 are shown in Fig. 1. High CO2 emissions were found in southwest states, northeastern coasts, and Florida. Table 2 provides the Spearman's rank correlation coefficients (rho) between each covariate and the two outcomes (change in diabetes and obesity prevalence). The Spearman's rank correlation demonstrated evidence that counties with higher changes in obesity generally had higher changes in diabetes though the two were not strongly correlated (rho = 0.12). Counties that had higher change in diabetes also had higher mean CO2 emissions (rho = 0.16). Change in obesity prevalence was weakly correlated with CO2 emission (rho = 0.03). Counties with increase in physical inactivity, population density, and percent black or African American all showed increasing change in prevalence of diabetes and obesity, while education attainment, percent white, and median age showed a decrease in prevalence of both diseases. In models adjusted for sociodemographic variables, indicators of urbanicity, and spatial autocorrelation (exponential), higher CO2 emission levels were marginally associated with larger changes in obesity prevalence from 2004 to 2008 (p for trend = 0.06) (Table 3). As compared to counties within the first quintile, the upper four quintiles had 0.093% (95% CI, − 0.115% to 0.300%), 0.226% (0.007% to 0.445%), 0.163% (− 0.073% to 0.399%), and 0.241% (− 0.014% to 0.497%) larger changes in prevalence than the first quintile. The strength of the association slightly decreased and became nonsignificant with further adjustment for PM2.5 (changes in obesity prevalence in the upper four quintiles: 0.068% (95% CI, − 0.140% to 0.276%), 0.171% (− 0.051% to 0.392%), 0.111% (− 0.133% to 0.354%), and 0.196% (− 0.071% to
Table 1 Distributions of the study variables for the 3019 counties analyzed. Mean (SD)
Average CO2 emission (million tons emitted) Prevalence of obesity in 2004 (%) Prevalence of obesity in 2008 (%) Change in obesity (%) Prevalence of diabetes in 2004 (%) Prevalence of diabetes in 2008 (%) Change in diabetes (%) Percent malea Median agea Percent Hispanic or Latino Percent White Percent Black or African American Percent Asian Percent below poverty Percent of population 25 years and older with a high school degree or equivalency Population density (population per square mile) Percent physical inactivity in county Density for fast food restaurant (number of restaurants per 100,000 residents) Average PM2.5 concentrationb
Range
0.55 (1.29) 0.002, 21.7 25.3 (3.25) 12.3, 38.0 28.9 (3.71) 11.7, 43.7 3.67 (1.89) −3.40, 12.6 8.29 (1.59) 3.00, 14.6 9.93 (2.05) 3.00, 18.2 1.64 (1.67) −6.20, 9.50 49.8 (2.22) 42.7, 74.9 39.4 (4.82) 21.7, 55.0 7.45 (12.6) 0.00, 98.6 84.2 (16.1) 5.75, 100 8.76 (14.2) 0.00, 86.8 1.09 (2.46) 0.00, 47.5 15.4 (6.38) 1.40, 51.0 82.5 (7.54) 46.5, 98.7 232 (1745) 0.08, 71,000 26.4 (5.10) 9.52, 43.5 68.8 (36.1) 0.00, 867 11.0 (2.47) 4.48, 20.1
Quintile of CO2 emission [mean (SD)] Q1
Q2
Q3
Q4
Q5
0.03 (0.01) 25.4 (3.28) 29.1 (3.77) 3.50 (1.81) 8.51 (1.68) 10.1 (2.03) 1.20 (1.66) 50.1 (2.87) 42.3 (4.96) 5.90 (12.5) 87.3 (17.2) 6.77 (15.0) 0.30 (0.45) 16.9 (7.53) 80.9 (8.52) 17.2 (17.5) 28.0 (4.33) 59.0 (57.9) 9.63 (2.34)
0.06 (0.01) 25.6 (3.13) 29.1 (3.53) 3.66 (1.75) 8.33 (1.59) 9.96 (2.07) 1.41 (1.62) 50.0 (2.46) 40.5 (4.29) 7.04 (13.9) 86.0 (15.9) 8.26 (15.1) 0.45 (0.56) 16.4 (6.48) 80.6 (8.02) 37.3 (87.6) 27.5 (5.07) 63.3 (31.5) 10.7 (2.36)
0.13 (0.03) 25.5 (3.15) 29.2 (3.52) 3.83 (1.94) 8.31 (1.60) 10.1 (2.07) 1.68 (1.64) 49.7 (2.00) 39.0 (4.41) 6.61 (11.7) 84.2 (16.0) 9.71 (15.7) 0.63 (0.69) 15.9 (6.30) 81.7 (7.42) 64.5 (85.6) 26.9 (5.36) 69.4 (26.6) 11.1 (2.22)
0.31 (0.10) 25.2 (3.36) 28.9 (3.91) 3.72 (1.98) 8.21 (1.52) 9.93 (1.98) 1.99 (1.62) 49.7 (2.06) 37.8 (4.57) 7.16 (10.9) 84.7 (14.1) 7.94 (11.3) 1.30 (2.83) 14.1 (5.38) 84.3 (6.35) 137 (138) 25.1 (4.87) 75.9 (24.2) 11.5 (2.31)
2.23 (2.19) 24.8 (3.32) 28.4 (3.89) 3.61 (1.95) 8.05 (1.51) 9.70 (2.09) 1.92 (1.68) 49.3 (1.34) 37.2 (4.01) 10.5 (13.5) 78.8 (15.9) 11.1 (13.2) 2.73 (4.14) 13.7 (5.29) 84.8 (6.06) 908 (3828) 24.5 (4.91) 81.4 (23.3) 12.1 (2.39)
All variables had statistically significant linear trends across the quintiles of CO2 emission (p for trend b0.0001) except obesity variables. a These variables were based on all age groups within each county, as there is no available data solely on adults in the U.S. American Community Survey. b PM2.5 available in 2994 counties (no PM2.5 data available in Alaska and Hawaii).
114
A.R. Zheutlin et al. / Environment International 73 (2014) 111–116
Fig. 2 shows the temporal trends in biannual averages of atmospheric CO2 concentrations at Mauna Loa, Hawaii, and serum bicarbonate concentrations in U.S. adults. There was an increasing trend in serum bicarbonate concentrations. On average, a 0.27 mmol/L (95% CI, 0.20 to 0.33) increased every cycle (every two years) between 1999/2000 and 2011/ 2012 (data not shown). 4. Discussion
Fig. 1. Maps of the United States demonstrating county-levels of CO2 emissions (average of annual levels between 2004 and 2008; million tons of carbon) (A), change in the prevalence (%) of obesity (B) and diagnosed diabetes (C) from 2004 to 2008.
0.463%); p for trend = 0.17). For diabetes, we found a significant positive association between CO2 emission and changes in diabetes prevalence (changes in diabetes prevalence in the upper four quintiles: − 0.086% (95% CI, − 0.255% to 0.083%), 0.119% (− 0.159% to 0.197%), 0.156% (− 0.036% to 0.348%), and 0.121% (− 0.087% to 0.328%); p for trend b 0.05) but the association became null after controlling for PM2.5 (changes in diabetes prevalence in the upper four quintiles: − 0.136% (95% CI, −0.309% to 0.037%), −0.044% (− 0.225% to 0.138%), 0.087% (− 0.109% to 0.283%), and − 0.027% (− 0.242% to 0.187%); p-trend = 0.49).
Table 2 Spearman correlation coefficients between covariates considered and the outcomes (changes in the prevalence of obesity and diabetes from 2004 to 2008) (N = 3019).
Change in obesity Change in diabetes Average CO2 emission Percent male Median age Percent Hispanic or Latino Percent White Percent Black or African American Percent Asian Percent below poverty Percent of population 25 years and older with a high school degree or equivalency Population density Percent physical inactivity Density of fast food restaurant Average PM2.5 concentration
Change in obesity
Change in diabetes
1 0.12 0.03 −0.02 −0.11 −0.06 −0.03 0.08 −0.02 0.09 −0.08
0.12 1 0.16 0.05 −0.34 0.07 −0.06 0.16 0.14 −0.02 −0.05
0.05 0.15 −0.07 0.11
0.28 0.05 0.06 0.26
Statistically non-significant correlations (p-value N 0.05) were between change in obesity and CO2 emission; percent male; percent white; and percent Asian; and between change in diabetes and percent below poverty. p-values for all other correlation coefficients were b0.01.
This study does not support the hypothesis that CO2, a leading driver of climate change, may be linked to increasing trends in obesity and diabetes. Although statistical significance was not achieved, the association between CO2 emission and the change in obesity prevalence showed the expected positive direction even after controlling for ambient PM2.5. On the other hand, the significant association between CO2 emission and the change in diabetes prevalence completely disappeared after controlling for ambient PM2.5, suggesting that fine particles, not CO2, may play a role in the increasing trend of diabetes prevalence and are an important confounder of the association between CO2 and the change in diabetes prevalence. The present study also showed that serum bicarbonate, a marker of body CO2 levels, increased significantly in the general U.S. population over the past 14 years. Potential factors that may have driven this trend including atmospheric CO2 were not determined. Our findings suggest that increases in sources of CO2 from fossil fuel combustion in the U.S. may not only have implications for global warming but could also be a contributing factor to the rising rates of obesity prevalence. This is consistent with a biological model proposed by Hersoug et al. (2012), who suggest that serum CO2 can be a catalyst for weight gain, although our study findings cannot be interpreted as causal. Our findings should be interpreted with caution. The nature of this study does not allow for conclusion that increased CO2 exposure shows a causal relationship with diabetes or obesity. This is an ecological study and therefore may be the result of the ecological fallacy (Piantadosi et al., 1988). For example, areas of greater CO2 emissions may have greater average income and/or greater use of vehicles. Bassetti et al. has previously demonstrated a “strong, long-run, positive relationship between income and CO2 emissions” (Bassetti et al., 2013). Health care access may be greater in those areas and have higher diagnosis rates because of this. Also, CO2 emissions from motor vehicles contribute to approximately 31% of total U.S. CO2 releases and 26% of total U.S. greenhouse gas emissions in 2011 (EPA, 2013b). Although we controlled for neighborhood indicators of socioeconomic position (e.g., race/ethnicity, educational attainment, and percent below poverty), population density, physical inactivity, access to fast food restaurants, and ambient fine particles, it remains possible that there is residual confounding by other factors unknown to this study. We also cannot rule out a possibility of reverse causality, in that, counties with increased obesity prevalence emit more CO2. Some argue that obese individuals may contribute to CO2 emission through increased food consumption and greater requirements for transportation energy (Edwards and Roberts, 2009). Edwards and Roberts (2009) reported that obese individuals lead to an increase of 0.173 gigatons and 2.038 gigatons in greenhouse gas emissions in the U.S. per year due to greater energy requirements for car and air travel. They also suggested that greater availability of cheap marketable food is connected to higher rates of purchase without complete consumption, which increases food waste and leads to more greenhouse gas emissions when it decomposes (Edwards and Roberts, 2009). In 2003 the energy burden of wasted food amounted to 4% of the total oil consumption in the United States (Hall et al., 2009). While many have studied obesity as a cause for CO2, evidence to support this is not convincing. Obesity is much more common in low socioeconomic areas, and the availability for personal automobile utilization is significantly lower in these areas (Gallar, 2010). Additionally, no study of CO2 emissions has been able to identify purchasing groups, and to assume that it is obese individuals consuming goods that
A.R. Zheutlin et al. / Environment International 73 (2014) 111–116
115
Table 3 Mean difference in change (95% CI) in obesity and diabetes prevalence between 2004 and 2008 for each quintile of CO2 emissions. Quintile of CO2 emission (range, million tons)
Obesity Crudea
p for trend
AIC
Q1 b0.0414
Q2 0.0414–0.0881
Q3 0.0881–0.1810
Q4 0.1810–0.5540
Q5 N0.5540
0
0.099 (−0.108,0.306) 0.093 (−0.115,0.300) 0.068 (−0.140, 0.276)
0.242 (0.031,0.453) 0.226 (0.007,0.445) 0.171 (−0.051, 0.392)
0.130 (−0.088,0.349) 0.163 (−0.073,0.399) 0.111 (−0.133, 0.354)
0.145 (−0.080,0.370) 0.241 (−0.014,0.497) 0.196 (−0.071, 0.463)
0.21
12,298.6
0.06
12,302.3
0.17
12,158.0
0.126 (−0.052,0.304) −0.086 (−0.255,0.083) −0.136 (−0.309, 0.037)
0.390 (0.208,0.571) 0.019 (−0.159,0.197) −0.044 (−0.225, 0.138)
0.658 (0.470,0.845) 0.156 (−0.036,0.348) 0.087 (−0.109, 0.283)
0.600 (0.407,0.793) 0.121 (−0.087,0.328) −0.027 (−0.242, 0.187)
b0.0001
11,386.5
0.05
11,066.6
0.49
11,054.5
b
Model 1
0
Model 2c
0
Diabetes Crudea
0
Model 1b
0
Model 2c
0
AIC, Akaike Information Criteria. a Crude: considering only spatial correlation (exponential correlation structure). b Model 1: adjusted for change in obesity prevalence (diabetes only), percent male, median age, percent Hispanic or Latino, percent white, percent black or African American, percent Asian, percent below poverty, percent of population 25 years and older with a high school degree or equivalency, population density, percent physical inactivity, density for fast food restaurant, as well as exponential spatial correlation. c Model 2: model 1 + PM2.5.
contribute to greenhouse gases is not based upon data (Gallar, 2010). Nevertheless, further investigation is needed to identify temporality of the association between CO2 emission and obesity. We examined CO2 emissions rather than atmospheric CO2 concentrations as the primary exposure variable because data for atmospheric CO2 are restricted to few testing stations that do not provide an adequate representation of county-level concentrations. Although atmospheric CO2 concentrations are known to be correlated with CO2 emissions (IPCC, 2013), it is unclear to what extent spatial variations in atmospheric CO2 concentrations exist at the county-level. Our statistical approaches to control for spatial autocorrelations and categorization of exposure could account for this potential issue though this study is still subject to exposure misclassification. Despite many limitations and non-significant findings, this study showed an indication of a possible link between CO2 from fossil fuel combustion and the prevalence of obesity. However, underlying biological mechanisms are poorly understood. It is hypothesized that increased ambient CO2 exposure can lower blood pH which in turn initiates a cascade response in the hypothalamus by activating orexin neurons (Hersoug et al., 2012). Recent studies have shown possible mechanisms by which orexin can impact weight disposition of organisms. Under hypercapnic conditions of 10% CO2, Sunanaga et al. (2009) measured increased stimulation of orexin neurons by CO2 and H+ demonstrated by elevated c-Fos expression. Both CO2 and extracellular acid are
physiochemical signals for wakefulness, and excess presence in serum can lead to higher excitability (Williams et al., 2007). Longer exposures to elevated CO2 can result in decreased pH of the blood, even after exposure to air with lower concentration of CO2 (King et al., 1955). This increased acidity of the blood can restrict movement of ions through post-synaptic K+ leak-like channels, which can increase the firing of orexin neurons, and elevate the concentration of orexin neurotransmitters in the body (Williams et al., 2007). By stimulating the body's desire for increased caloric intake, individuals are disposed to weight gain. The alteration of homeostatic regulation by the hypothalamus, leading to increase orexin neuron activity, induced by CO2, is a plausible path for excess weight gain (Hersoug et al., 2012). Further studies are necessary to elucidate potential biological mechanisms of the link between CO2 exposure and weight gain and obesity. In particular, the increasing trend in serum bicarbonate found in our study is related to elevated ambient CO2 and climate change needs to be determined. Our study suggests that fine particles, not CO2 emission, may play a role in the change in diabetes prevalence. In the fully adjusted model including both CO2 emission and ambient PM2.5, a 10 μg/m3 increase in PM2.5 was associated with a 0.90% (95% CI, 0.58% to 1.21%) increase in the diabetes prevalence (p b 0.0001, data not shown). PM2.5 was not associated with changes in obesity prevalence (p = 0.26, data not shown). This is consistent with the previous finding that counties with high levels of PM2.5 had significantly greater prevalence of diagnosed
Fig. 2. Temporal trends in biannual averages of atmospheric CO2 concentrations at Mauna Loa, Hawaii, and serum bicarbonate concentrations in U.S. adults.
116
A.R. Zheutlin et al. / Environment International 73 (2014) 111–116
diabetes in the U.S. (Pearson et al., 2010). Numerous observational studies also supported the link between long-term air pollution exposure to the risk of type-2 diabetes (Brook et al., 2013; Coogan et al., 2012; Kramer et al., 2010). PM2.5 exposure to mice, with diet-induced insulin resistance, promoted greater insulin resistance and visceral inflammation and influenced insulin signaling pathway and apoptosis, key biological mechanisms in type-2 diabetes (Sun et al., 2009; X. Xu et al., 2011; Z. Xu et al., 2011). Although obesity is an important risk factor of type-2 diabetes (Schulze and Hu, 2005) and thus we hypothesized that CO2 may influence two related metabolic outcomes, our study does not support this hypothesis probably due to different biological mechanisms between CO2 and PM2.5. The present study examined data between 2004 and 2008 because CO2 emission data in the U.S. at the county level are available only in this period and it allowed for maximum county inclusion from all of the providing data sources. It may be that five years does not provide a long enough time frame to fully examine how CO2 levels play into obesity or diabetes prevalence at the population level. Measuring CO2, especially as ambient concentrations, over an extended period of time has been restricted to specific sites globally, and is not available on a county level in the U.S. However, data from the Mauna Loa observatory demonstrates the increase in ambient CO2. Specifically, from 2004 to 2008 atmospheric CO2 from this location increased from 377.49 to 385.85 ppm and continued to increase to 396.48 ppm in 2013 (NOAA/ESRL, 2014). In summary, our study does not support a link between CO2 emissions from fossil fuel combustion and increasing trends of obesity and diabetes, though there was an indication of a possible link between CO2 and obesity. Potential health impacts of climate change have received huge attentions recently, especially following acute natural disasters, such as Hurricane Sandy in 2012 and the Oklahoma tornadoes in 2013. Increased CO2 is a key component of global climate change, which poses numerous threats such as food security by altering the preferred growing microclimate, human safety by heightened intensity and frequency of natural disasters (Fischer et al., 1994; Mills, 2009). However, the subtle accumulative effects of increased CO2 from climate change on human health are relatively unknown. It is an urgent need to explore potential cumulative effects of climate change, such as increasing atmospheric CO2, especially on homeostatic regulation given the widespread exposure to CO2 in large populations. Supplementary data to this article can be found online at http://dx. doi.org/10.1016/j.envint.2014.07.012. Conflict of interest There are no relevant conflicts of interest to disclose. Acknowledgment This study was supported by the National Institute of Environmental Health Sciences (NIEHS) grant K01-ES016587. Author contributions: Study concept and design: Park. Acquisition of data: Zheutlin and Park. Analysis and interpretation of data: Zheutlin and Park. Drafting of the manuscript: Zheutlin. Critical revision of the manuscript for important intellectual content: Adar and Park. Statistical analysis: Zheutlin and Park. References ASPE. 2014 poverty guidelines. Washington, DC: U.S. Department of Health & Human Services, Office of the Assistant Secretary for Planning and Evaluation; 2014. Bassetti T, Benos N, Karagiannis S. CO2 emissions and income dynamics: what does the global evidence tell us? Environ Resour Econ 2013;54:101–25.
Brook RD, Cakmak S, Turner MC, Brook JR, Crouse DL, Peters PA, et al. Long-term fine particulate matter exposure and mortality from diabetes in Canada. Diabetes Care 2013;36:3313–20. CDC. National diabetes fact sheet: national estimates and general information on diabetes and prediabetes in the United States, 2011. Atlanta, GA: U.S. Department of Health and Human Services, Centers for Disease Control and Prevention; 2011. CDC. Behavioral risk factor surveillance system. Atlanta, GA: Centers for Disease Control and Prevention; 2012a. CDC. Diabetes interactive atlases. Atlanta, GA: Centers for Disease Control and Prevention; 2012b. CDC. Diabetes data & trends: frequently asked questions (FAQ). Atlanta, GA: Centers for Disease Control and Prevention; 2013a. CDC. National Health and Nutrition Examination Survey Data. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention, National Center for Health Statistics; 2013b. CDC. Outdoor air: monitor + model air data. Atlanta, GA: Centers for Disease Control and Prevention; 2013c. Chen L, Magliano DJ, Zimmet PZ. The worldwide epidemiology of type 2 diabetes mellitus —present and future perspectives. Nat Rev Endocrinol 2012;8:228–36. Coogan PF, White LF, Jerrett M, Brook RD, Su JG, Seto E, et al. Air pollution and incidence of hypertension and diabetes mellitus in black women living in Los Angeles. Circulation 2012;125:767–72. Department of Commerce. American FactFinder. Washington, DC: U.S. Department of Commerce; 2012. Edwards P, Roberts I. Population adiposity and climate change. Int J Epidemiol 2009;38: 1137–40. EPA. Inventory of U.S. greenhouse gas emissions and sinks: 1990–2011. EPA 430-R-13001, Washington, DC: U.S. Environmental Protection Agency; 2013a. EPA. Overview of greenhouse gases: carbon dioxide emissions. Washington, DC: U.S. Environmental Protection Agency; 2013b. Fischer G, Frohberg K, Parry ML, Rosenzweig C. Climate change and world food supply, demand and trade: who benefits, who loses? Glob Environ Chang 1994; 4:7–23. Gallar M. Obesity and climate change. Int J Epidemiol 2010;39:1398–9. Gurney KR, Mendoza DL, Zhou Y, Fischer ML, Miller CC, Geethakumar S, et al. High resolution fossil fuel combustion CO2 emission fluxes for the United States. Environ Sci Technol 2009;43:5535–41. Hall KD, Guo J, Dore M, Chow CC. The progressive increase of food waste in America and its environmental impact. PLoS One 2009;4:e7940. Hersoug LG, Sjodin A, Astrup A. A proposed potential role for increasing atmospheric CO2 as a promoter of weight gain and obesity. Nutr Diabetes 2012;2:e31. Hoven AD. Obesity as a disease? New York City, NY: Huffington Post; 2013. IPCC. Carbon dioxide: projected emissions and concentrations. Intergovernmental panel on climate change; 2013. King CT, Williams EE, Mego JL, Schaefer KE. Adrenal function during prolonged exposure to low concentration of carbon dioxide. Am J Physiol 1955;183:46–52. Klimentidis YC, Beasley TM, Lin HY, Murati G, Glass GE, Guyton M, et al. Canaries in the coal mine: a cross-species analysis of the plurality of obesity epidemics. Proc Biol Sci 2011;278:1626–32. Kramer U, Herder C, Sugiri D, Strassburger K, Schikowski T, Ranft U, et al. Traffic-related air pollution and incident type 2 diabetes: results from the SALIA cohort study. Environ Health Perspect 2010;118:1273–9. Malik VS, Willett WC, Hu FB. Global obesity: trends, risk factors and policy implications. Nat Rev Endocrinol 2013;9:13–27. Mills DM. Climate change, extreme weather events, and us health impacts: what can we say? J Occup Environ Med 2009;51:26–32. NOAA/ESRL. Trends in atmospheric carbon dioxide. U.S. Department of Commerce, National Oceanic and Atmospheric Administration, Earth System Research Laboratory; 2014. Pearson JF, Bachireddy C, Shyamprasad S, Goldfine AB, Brownstein JS. Association between fine particulate matter and diabetes prevalence in the U.S. Diabetes Care 2010;33:2196–201. Piantadosi S, Byar DP, Green SB. The ecological fallacy. Am J Epidemiol 1988;127: 893–904. Schulze MB, Hu FB. Primary prevention of diabetes: what can be done and how much can be prevented? Annu Rev Public Health 2005;26:445–67. Sun Q, Yue P, Deiuliis JA, Lumeng CN, Kampfrath T, Mikolaj MB, et al. Ambient air pollution exaggerates adipose inflammation and insulin resistance in a mouse model of diet-induced obesity. Circulation 2009;119:538–46. Sunanaga J, Deng BS, Zhang W, Kanmura Y, Kuwaki T. CO2 activates orexin-containing neurons in mice. Respir Physiol Neurobiol 2009;166:184–6. Thayer KA, Heindel JJ, Bucher JR, Gallo MA. Role of environmental chemicals in diabetes and obesity: a National Toxicology Program workshop review. Environ Health Perspect 2012;120:779–89. Williams RH, Jensen LT, Verkhratsky A, Fugger L, Burdakov D. Control of hypothalamic orexin neurons by acid and CO2. Proc Natl Acad Sci U S A 2007;104:10685–90. Xu X, Liu C, Xu Z, Tzan K, Zhong M, Wang A, et al. Long-term exposure to ambient fine particulate pollution induces insulin resistance and mitochondrial alteration in adipose tissue. Toxicol Sci 2011a;124:88–98. Xu Z, Xu X, Zhong M, Hotchkiss IP, Lewandowski RP, Wagner JG, et al. Ambient particulate air pollution induces oxidative stress and alterations of mitochondria and gene expression in brown and white adipose tissues. Part Fibre Toxicol 2011b;8:20.